Skip to content

Documents ​

Documents provide model-readable text or files alongside a request. Use them for small, relevant inputs—not as a replacement for retrieval over a large corpus.

Add text documents to a completion ​

ts
const result = await createCompletion(model, {
  input: 'What changed in this release?',
  documents: [
    {
      id: 'release-notes',
      text: releaseNotes,
    },
  ],
})

This form is useful when application code already has the text and wants to name the document for one request.

Add a file to a user message ​

ts
import { Message, UserContent } from '@anvia/core'

const message = Message.user([
  UserContent.text('Summarize this report.'),
  UserContent.documentBase64(pdfBase64, 'application/pdf', {
    filename: 'incident-report.pdf',
  }),
])

Use documentUrl(...) for an accessible URL or documentBase64(...) for in-memory file content. Provider limits and supported media types vary.

Choose documents or retrieval ​

SourceUse
One small document already selected by the applicationRequest document
Many files requiring parsing or chunkingLoader and ingestion flow
A large or changing knowledge collectionKnowledge retrieval

Validate file type, size, ownership, and access before content reaches the model. Do not rely on prompt instructions to hide unauthorized documents.

Built for Anvia.